Mathematical Reasoning on AIME 2024 (Acc, Len, Red)
76.3AccuracyOPSDC
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OPSDCModel Architecture=Qwen3-14B, Token Budget=30K, Training Protocol=periodic teacher update (M=50)2026.03 | 76.3 | 7,577 | 41 | |
| Base ModelModel Architecture=Qwen3-8B, Token Budget=30K2026.03 | 72.5 | 14,170 | — | |
| OPSDCModel Architecture=Qwen3-8B, Token Budget=30K, Training Protocol=periodic teacher update (M=50)2026.03 | 69.6 | 9,152 | 35.4 | |
| Concise promptModel Architecture=Qwen3-14B, Token Budget=30K, Prompt Strategy=conciseness instruction at inference2026.03 | 68.3 | 9,866 | 23.2 | |
| Concise promptModel Architecture=Qwen3-8B, Token Budget=30K, Prompt Strategy=conciseness instruction at inference2026.03 | 67.5 | 11,589 | 18.2 | |
| Base ModelModel Architecture=Qwen3-14B, Token Budget=30K2026.03 | 65.8 | 12,844 | — |